Advanced Visualisation (Made Easy) in R Workshop
T.J. McKinley (t.mckinley@exeter.ac.uk)
Workshop notes
‘Tidy’ data
“Tidy datasets are all alike but every messy dataset is messy in its own way.”—Hadley Wickham
Specifically, a tidy data set is one in which:
- rows contain different observations;
- columns contain different variables†;
- cells contain values.
The idea of ‘tidy’ data gives rise to the nomenclature of the
tidyverse.
†but variables have to make sense, so having a
Sex column with levels: Male,
Female, All would not make sense because
Male and Female are nested within
All
The tidyverse
The tidyverse
is a suite of R packages designed for data science. All
packages share an underlying design philosophy and grammar. All exploit
the power of ‘tidy’ datasets.
Rather than load each separate package, we will load them all† using:
library(tidyverse)†they can be installed using
install.packages("tidyverse"), though this takes some
time…
Further reading
I would highly recommend:
What we’re aiming for…
Example: Titanic survival data
Let’s have a look at some data:
On April 15, 1912, during her maiden voyage, the Titanic sank after colliding with an iceberg, killing 1502 out of 2224 passengers and crew.
This tragedy shocked the international community and led to better safety regulations for ships.
Introduction to ggplot2
For this workshop we will assume that we have a ‘tidy’ data set that
we wish to visualise, and we will use the package
ggplot2† to produce beautiful plots of various
different types from this data set.
ggplot2 is a hugely powerful package, and is the de
facto package of choice for many data scientists. We will see how
this package makes complex plots easy!
†loaded automatically as part of tidyverse
Example: Titanic survival data
For comparison let’s look at standard base R, next to
ggplot2:
barplot(table(titanic$survived))
ggplot(titanic) +
geom_bar(aes(x = survived))
Example: Titanic survival data
Let’s explore differences in survival between men and women first.
barplot(table(titanic$gender,
titanic$survived))
ggplot(titanic) +
geom_bar(aes(x = survived,
fill = gender))
Example: Titanic survival data
Now let’s add passenger class into the mix…
par(mfrow = c(1, 3))
for(i in unique(titanic$pclass)) {
temp <- titanic[
titanic$pclass == i, ]
barplot(table(temp$gender,
temp$survived))
}
ggplot(titanic) +
geom_bar(aes(x = survived,
fill = gender)) +
facet_wrap(~pclass)
Example: Titanic survival data
Full base R solution:
Example: Titanic survival data
Compare this to the ggplot2
version:
ggplot(titanic) +
geom_bar(aes(x = survived, fill = gender)) +
facet_wrap(~pclass)
Data
Firstly, ggplot2 always
operates on data.frame† objects. Hence we always
set up a plot by telling the ggplot() function which data
frame we are operating on e.g.
ggplot(titanic)†or tibble objects—which are special types of
data frame that you might come across if you delve more into
tidyverse
Building plots
To stratify by gender is easy. We simply set the
corresponding aesthetic: in this case we can use the
fill aesthetic to produce different bars for each level of
the fill variable:
ggplot(titanic) +
geom_bar(aes(x = survived,
fill = gender))
Facets
ggplot(titanic) +
geom_bar(aes(x = survived,
fill = gender)) +
facet_wrap(~pclass)
Facets
We may want to swap variables around, which is now straightforward:
ggplot(titanic) +
geom_bar(aes(x = survived,
fill = pclass)) +
facet_wrap(~gender)
Putting these ideas together
facet_grid() is used less often, but allows you to facet
by more than one variable e.g.
ggplot(titanic) +
geom_bar(aes(x = survived)) +
facet_grid(pclass ~ gender)
Example: Titanic survival data
Let’s see whether there is any relationship between age
and fare.
ggplot(titanic) +
geom_point(aes(x = fare, y = age))
Example: Titanic survival data
It may be easier to visualise if we add a smoothed line-of-best-fit (with global aesthetics):
ggplot(titanic, aes(x = fare, y = age)) +
geom_point() +
stat_smooth()You can change the default parameters of the smoother, such as the
bandwidth (see ?stat_smooth).
Useful for exploratory purposes, but no substitute for a proper model!
Example: Titanic survival data
Not much going on here in my opinion. Note that stat_*()
operations preserve aesthetics and facets. For example:
ggplot(titanic,
aes(
x = fare,
y = age,
colour = survived
)
) +
geom_point() +
facet_wrap(~gender) +
stat_smooth()Notice smoother is heavily influenced by outliers here (producing large uncertainty bounds).
Scales
Scales control the details of how data values are translated to visual properties.
These allow us to control translations from data to
aesthetics (e.g. change default colour schemes etc.). Scales are
generally of the form scale_AESTHETIC_*() e.g.
ggplot(titanic) +
geom_bar(
aes(x = survived, fill = gender)
) +
scale_fill_manual(
values = c("red", "blue")
)
Positions
Some geoms offer alternative positions. For example,
to change from a stacked bar plot to a
side-by-side bar plot, we can set the
position argument of geom_bar() to
"dodge" e.g.
ggplot(titanic) +
geom_bar(
aes(
x = survived,
fill = gender
),
position = "dodge"
)